Amazon is one of the largest e-commerce companies in the world, but it entered the cloud business with a storm in 2006 ^[1], thanks in part to the expertise it had developed as an Internet retail giant. Amazon's cloud business operates as a separate divison called AWS and in 2015 they reached $14 billion in sales ^[2].

Amazon offers multiple types of cloud storage combined with 70+ other services. So to make it easier to compare Amazon's cloud storage options to other cloud storage providers, the following sections classify Amazon's cloud storage options into two major groups: file storage -- which includes photo & video -- and application cloud storage.

Amazon file cloud storage

Amazon file cloud storage is offered in two paid plans under the name Amazon Drive ^[3]. The first plan is coupled with Amazon's frequent customer program Amazon Prime for which you pay $10.99 (USD) a month or $99 (USD) a year, in which case you also a get a series of Amazon perks (e.g. free two-day shipping, free access to select movies, TV shows, music & ebooks). The second plan is a standalone cloud file storage service priced at an average of $59.99 (USD) per Tera Byte per year.

Amazon Drive unlimited limits: Storage, bandwidth & file types

Both Amazon Drive plans use the term unlimited in their marketing material, but there's a lot of legalese behind the term. Let's start with what the actual plans offer and then explore what can be implied by unlimited in the terms of service.

• Amazon Prime photos ^[4].- Unlimited photo storage & 5 GB of storage for videos, music and other files, for $10.99 (USD) a month or $99 (USD) a year.
• Amazon Drive storage.- File storage for photos, videos, music, documents & more for: $11.99 (USD) a year for 100 GB; $59.99 (USD) a year for 1 TB, up to 10 TB in 1 TB increments for $59.99 (USD) each TB; as well as, a 20 TB and 30 TB plan for $1199.80 (USD) and $1799.70 (USD) a year, respectively, with each TB also costing $59.99 (USD) per year.^[5]

The only conclusive fact about limits with Amazon Drive is the Prime Photos plan offers 5 GB of storage for videos, music and other file types. But the unlimited term in both Amazon Cloud Drive plans, is in fact seriously limited on various fronts.

• Non-commercial use.- In section 1.2 of the terms of service^[6], it states: You may use the Services only to store, retrieve, manage, organize, and access Your Files for personal, non-commercial purposes using the features and functionality we make available. You may not use the Services to store, transfer, or distribute content of or on behalf of third parties, to operate your own file storage application or service, to operate a photography business or other commercial service, or to resell any part of the Services.
• File types.- In section 3.2 of the terms of service^[7], it states: There may be limits on the types of content you can store and share using the Services, such as file types we do not support, and on the number or type of devices you can use to access the Services. We may impose other restrictions on use of the Services.
• Do not exceed normal use.- In section 5.2 of the terms of service^[8], it states: We may terminate the Agreement or restrict, suspend, or terminate your use of the Services at our discretion without notice at any time, including if we determine that your use violates the Agreement, is improper, substantially exceeds or differs from normal use by other users.
• Potential loss of files if restricted/suspended/terminated.- In the same section 5.2 of the terms of service^[8], it also states: If your Service Plan is restricted, suspended, or terminated, you may be unable to access Your Files and you will not receive any refund of fees or any other compensation.

Like many legal statements, the pitfall is all of these terms are subject to interpretation. In this sense, Amazon Cloud Drive has one of the most obscure terms of any cloud storage provider. Unfortuantly, because there are no answers in plain English regarding limits, you should assume the worse and avoid activities that can resemble term violations. As a rule of thumb, think twice before using Amazon Prime photos or Amazon Drive storage, if you think you could fall into:

• Fuzzy boundaries in work-personal activity.- If you go to a corporate party and share photos with colleagues, you could get flagged because of 'non-commercial' activity on corporate IP addresses or something along these lines.
• Specialized or non-standard file formats.- If you backup a personal database or use an advanced photo format (e.g. RAW ^[8.5]), it might not be storable because it isn't a mainstream format or it's considered a 'professional' business format.
• Using more than industry average limits.- If you're looking to migrate from another cloud storage provider to cut costs and take advantage of Amazon's Cloud Drive 'unlimited' features, tread carefully. Assume anything beyond industry cloud storage averages (e.g. 100 Giga Bytes (GB) of daily bandwidth) can be construed as abnormal and put your account plan at risk.
• Accessibility of important files.- Although this applies to all cloud storage providers in general, this is particularly important in this case because the terms state you may be unable to access your files if the account is restricted, suspended or terminated -- caveat emptor.

But assuming you can get by with these limits, let's take one final look and see what Amazon Cloud Drive's price points can get you with the competition.

• Amazon Prime photos.- This requires an apples to oranges comparison, since the Prime photos plan is lumped together with other Amazon perks to make up an Amazon prime membership. If you're a heavy Amazon shopper, the price of the Amazon Prime membership alone can very well be justified by some of the other perks (e.g. free two-day shipping, free access to select movies, TV shows, music & ebooks) irrespective of file storage, so it's difficult to compare this plan's 'free' cloud storage since it's part of a larger plan that goes beyond cloud storage.
• Amazon Drive storage.- This plan starts at $11.99 (USD) a year for the first 100 GB and $59.99 (USD) a year for each subsequent TB up to a maximum of 30 TB. The Amazon Drive 100 GB plan comes to $0.99 (USD) per month for 100 GB, which makes it among the most competitive in its range vs. Google Drive for 100 GB at $1.99 (USD) a month vs. Apple iCloud for 50 GB at $0.99 (USD) vs. Microsoft OneDrive Basic for 50 GB at $1.99 (USD). The remaining Amazon Drive plans at $59.99 (USD) per TB a year, come to $4.99 (USD) per month per TB, which also place the plans in a very competitive range along the lines of either a PCloud Premium account or a Sync personal account.

Amazon application cloud storage

Amazon offers over 70+ cloud services for applications under the name AWS ^[9]. Each of these cloud services is priced using a wide variety of metrics, including: per hour, per request and per storage unit, among other units. Although obtaining an exact price for a given combination of cloud services can be time consuming, Amazon does offer a clear cut pricing tier if you're starting to use its application cloud.

• AWS free tier^[10].- An introductory 12-month trial to basic services, including : A 1GB RAM server (EC2 T2.micro) with 30GB of storage (EBS); 5 GB of object storage (S3), designed to easily store/retrieve files on the web; a 20 GB database instance (RDS db.t2.micro) from a major brand (e.g. MySQL, PostgreSQL); among other selected services.

So what's the catch to it being free ? That applications consume more services and run for longer periods of time than the free tier limits. A 1GB RAM server and 30 GB of storage is apt for only the smallest of applications, so if you need to run applications on a server with more RAM or a higher storage capacity, you need to pay the difference from day one. Not to mention that after one year, you have to pay every resource, period.

Now that you know what you can get for free with Amazon's application cloud services, it's important to understand the different product terms (e.g. EC2, S3, RDS) to be able to easily compare each service to other application cloud services offered by competitors like Google and Microsoft.

AWS servers

The main building block to run applications on the cloud are servers. But in AWS like other application cloud providers, servers are not sold as monolithic machines like a laptop or workstation that comes with a fixed amount of memory (RAM), processor (CPU) and storage (TBs of disk space).

In AWS, servers can be purchased on demand with a great deal of configuration options. This means that if your applications require more or less memory/processing power/storage, you can make the necessary adjustments in a few clicks to better suit your needs and budget.

Servers in AWS are named EC2 a short-handed notation for Elastic Compute Cloud. It's essential you first understand the parts that make up an AWS server, because the differences between these parts is the reason there are over 40 different types of AWS servers to choose from. The following figure illustrates the basic components of an AWS server, followed by a list of important facts.

Now that you're familiar with the basic components of AWS servers, lets take a look at the different type of servers. AWS EC2 servers are grouped into thirteen different types, with each instance type further sub-divided in up to six different models. Instance types use a letter-number notation (e.g. T2,R3,C4) to group servers by qualitative resources designed for things like graphic, storage or memory intensive tasks. Model classification uses a size notation (e.g.small,medium,large) to describe a server's quantitative resources like the amount of vCPUs and RAM capacity.

The following chart illustrates all the types of EC2 servers, classified by the amount of resources -- vCPUs and RAM -- they offer and their main purpose.

As you can see in this last chart, there are many servers with the same or similar amount of vCPUs and RAM, which makes using this criteria alone a weak choice. To narrow down the AWS server selection process, the previous chart also groups the more abstract instance types (e.g. T2,R3,C4) into five main purposes:

• General purpose (T2,M3,M4).- Servers that offer a balanced amount of resources for all types of applications.
• Compute optimized (C3,C4).- Servers that offer the most and best processing power for CPU bound applications. For example, higher amount of vCPUs, higher clock speeds, lower cost per vCPU.
• Memory optimized (R3,R4,X1).- Servers that offer the most and best memory capacity for RAM memory bound applications. For example, higher amount of RAM (GiB) and lower cost per GiB of RAM.
• Accelerated compute optimized for GPU (G2,P2) and FPGA (F1).- Servers that offer graphics processing unit (GPU) support for GPU bound applications, as well as servers that offer hardware acceleration with field programmable gate arrays (FPGAs) for applications that depend on acceleration code.
• Storage optimized (I3,D2).- Servers that offer the most and best storage capacity for storage/IO(Input/Output) bound applications. For example, higher amount of local SSD storage, lower cost per local GiB of storage.

Based on these purpose groups, you can limit AWS server options down to two or three instance types (i.e. letter-number types). But look at the previous chart once again and notice the pattern for servers in the same purpose group and their vCPUs & RAM. For example, the T2.large, M4.large and M3.large servers -- all in the general purpose group -- have 2 vCPUs and 8GiB of memory. Similarly, the C4.xlarge and C2.xlarge servers both have 4 vCPUs and 7.5GiB of memory and yet they're both optimized for processing. This pattern repeats itself across all groups, which means there are additional differences between AWS servers besides their main purpose, vCPUs an RAM.

The following table contains a series of additional AWS server features classified by purpose group to narrow down the server selection process to a single instance type (i.e. letter-number type).

Purpose group	General purpose			Compute optimized (Lowest CPU cost per unit)		Memory optimized (Lowest RAM cost per unit)			Accelerated compute optimized (GPU and FPGA support)			Storage optimized (Best disk IO w/ local SSD storage)
Feature	T2	M3	M4	C3	C4	R3	R4	X1	G2	P2	F1	I3	D2
Burstable CPU performance	✔
Dedicated EBS bandwidth included			✔		✔					✔			✔
Dedicated EBS bandwidth available at additional cost (May not available on all size types)		✔		✔		✔	✔	✔	✔		✔	✔
SSD instance storage		✔		✔		✔		✔	✔		✔	✔	✔
Cluster networking support			✔	✔	✔		✔	✔	✔	✔	✔	✔	✔

Server burstable CPU performance (Only applies to general purpose server selection)

Most servers don't use 100% of their processing power 24 hours a day for 7 days a week. Instead most server ebb and flow between 0% usage when there's no activity to 100+% when they hit peak loads. Because of this behavior and to offer more attractive pricing, AWS offers the general purpose T2 instance type with burstable CPU performance.

Burstable CPU performance means a server is allowed to burst into 100% CPU usage for a predetermined amount of time and is otherwise capped to a baseline CPU percentage. AWS determines the amount of burstable time it gives a server by means of CPU credits. A CPU credit unit represents 100% CPU use for one minute. The amount of CPU credits and baseline CPU percentage a server gets is determined by the size of a T2 instance type, illustrated in the following table:

Model	Baseline CPU %	CPU credits given per hour	Maximum earned CPU credits
T2.nano	5%	3 (3 minutes of 100% CPU every hour)	72 (1.2 hours of 100% CPU)
T2.micro	10%	6 (6 minutes of 100% CPU every hour)	144 (2.4 hours of 100% CPU)
T2.small	20%	12 (12 minutes of 100% CPU every hour)	288 (4.8 hours of 100% CPU)
T2.medium	40%	24 (24 minutes of 100% CPU every hour)	576 (9.6 hours of 100% CPU)
T2.large	60%	36 (36 minutes of 100% CPU every hour)	864 (14.4 hours of 100% CPU)
T2.xlarge	90%	54 (54 minutes of 100% CPU every hour)	1296 (21.6 hours of 100% CPU)
T2.2xlarge	135%	81 (81 minutes of 100% CPU every hour)	1944 (32.4 hours of 100% CPU)

• If a server never surpasses its baseline CPU % threshold, no CPU credits are used. If a server surpasses its baseline CPU % threshold, it uses CPU credits until the workload tapers and the server reaches its baseline CPU % again or all CPU credits are consumed. If no CPU credits are available, a server is capped to its baseline CPU % and performance may suffer with heavy CPU demands.
• CPU credits are given every hour and can be accumulated for up to a 24 hour worth balance, as shown in the maximum earned CPU credits column. If a server constantly remains with a high CPU credit balance, you can most likely switch to a smaller server as the current one is never reaching peak CPU usage. If a server constantly consumes its CPU credit balance, you should most likely switch to a larger server as the current one is constantly reaching peak CPU usage.
• Use the maximum earned CPU credits as an initial guideline to select a T2 instance type. For example, if you estimate a server workload will never exceed 100% CPU usage for more than 1.2 hours a day choose T2.nano, if you estimate a server workload will exceed 100% CPU usage for more than 12 hours every day but less than 14.4 hours a day choose T2.large. Think of maximum earned CPU credits as a car's gas tank, in this case it takes a full day to 'fill up' the tank (server) and you can burn through it as quickly or as slowly as you need, you'll just never be able to get more mileage than the tank (server) permits.

Server EBS bandwidth

As illustrated and described earlier, AWS servers store the bulk of their data (e.g. operating system, applications) on disk drives called EBS (Elastic Block Storage). However, as flexible as this architecture is to quickly upgrade/downgrade servers, EBS also has implications that influence server selection.

The communication between AWS EC2 servers and EBS disk drives takes place over a network, which means EBS operates as a type of network area storage (NAS). Although NAS technology offers benefits that have already been praised -- to quickly upgrade/downgrade servers -- its communication will never be as performant as having disk drives directly attached to a server, on top of which there will always be other traffic competing for the same network bandwidth used by an AWS server and EBS disk drives.

Because the communication between a server and its disk drives can be critical, AWS offers three options to configure communication between an EC2 server and EBS disk drives.

• No special EBS option.- The communication between an EC2 server and EBS disk drive gets no special treatment and the network bandwidth is shared with everything else that goes on with the server. This is the least performant EBS option, which is the only option available for the general purpose T2 instance type, as well as the default choice for the general purpose M3 instance type, compute optimized C3 instance type, memory optimized R3,R4 & X1 instance types, accelerated compute optimzied G2 & F1 instance types and the storage optimized I3 instance type.
• Dedicated EBS bandwidth at additional cost.- The communication between an EC2 server and EBS disk drive gets special treatment with dedicated bandwidth. The dedicated bandwidth available between an EC2 server and EBS disk drive depends on the size of the server (e.g.small,medium,large). You can pay for dedicated EBS bandwidth on the general purpose M3 instance type, compute optimized C3 instance type, memory optimized R3,R4 & X1 instance types, accelerated compute optimzied G2 & F1 instance types and the storage optimized I3 instance type.
• Dedicated EBS bandwidth included with the server.- The communication between an EC2 server and EBS disk drive gets special treatment with dedicated bandwidth and is included as part of the server cost. The dedicated bandwidth available between an EC2 server and EBS disk drive depends on the size of the server (e.g.small,medium,large). The following server instance types include dedicated EBS bandwidth with their cost: the general purpose M4 instance type, the compute optimized C4 instance type, the accelerated compute optimized P2 instance type and the storage optimized D2 instance type.

To provide you with additional criteria to see what exactly it's you get out of dedicated EBS bandwidth, the following table shows different performance levels for dedicated EBS bandwith.

Level	Max. bandwidth (Mbps)	Expected throughput (MB/s)	Max. IOPS (16 KB I/O size)
A	450	56.25	3600
B	500	62.5	4000
C	750	93.75	6000
D	1000	125	8000
E	2000	250	16000
F	4000	500	32000
G	5000	625	32500
H	10000	1250	65000

As you can see, the higher the dedicated EBS bandwidth between an EC2 server and EBS disk drive, the higher the guaranteed max bandwidth, expected throughput and Max. IOPS. It's important to note that you can't arbitrarily select an EBS bandwidth level for a server, each server type (e.g.small,medium,large) comes pre-assigned with an EBS bandwidth level, if you need more EBS bandwidth level you'll have to upgrade to a larger server. In addition, it's worth mentioning not all EBS bandwidth levels presented in the previous table are available across all instance types (e.g. you may need to change instance type to reach upper bound EBS bandwidth levels)

Server local storage

In certain circumstances, the performance metrics offered by EBS disk drives -- described in the previous table -- may not be enough to satisfy technical requirements. For example, if you want to run a database engine with quick response times or a caching layer, you want to use disk drives as close as possible to the server where it's run. In such cases you can use servers with local storage.

Local storage are disk drives that are directly attached to a server, which means they offer better disk performance than any EBS bandwidth optimized disk drive. The drawback of local SSD storage is the servers that offer this option, only have a fixed amount of this space (i.e. you can't increase/decrease space like with EBS disk drives), in addition, moving data from one server's local SSD storage to another requires a full disk copy, unlike EBS disk drives which can quickly be attached/detached across servers.

The amount and type of local storage on a server varies depending on a server's size and purpose group. For example, the purpose groups M3, C3, R3, X1 and G2 use Solid State Drive (SSD) to deliver high random I/O performance. Where as the I3 and F1 purpose groups use non-volatile memory express (NVMe) SSD which allows parallelism in SSD. The D2xlarge instance can offer one local storage drive, where as a 16xlarge instance can offer up to eighth local storage drives to parallelize and safeguard workloads with RAID. See the feature table at the bottomo of this page for additional details on local storage options.

Server pricing and time commitment

Once you decide on an AWS purpose group and server size based on technical requirements, you then need to decide on a time commitement to determine server pricing.